Marketing outcomes do not follow normal distributions. They follow power laws.
This matters more than most marketers realize.
In a normal distribution, average performance is meaningful. Optimization yields predictable returns. Spreading resources evenly is rational. In a power law distribution, averages are misleading. Most efforts produce negligible results. A small number of inputs generate the majority of outcomes.
If marketing returns follow power laws (and the evidence suggests they do), the standard playbook for budget allocation, channel diversification, keyword coverage, and campaign optimization is misaligned with how results actually accrue.
The Normal Distribution Assumption
Most marketing operations run on an implicit assumption: that performance is normally distributed across campaigns, keywords, channels, and audiences.
Under this model, each campaign contributes meaningfully to the whole. Some perform above average, some below, but all occupy a place on a bell curve clustered around a meaningful mean. The strategic implication is straightforward. Improve everything by small amounts, and the aggregate improves proportionally.
This assumption drives common practices:
"Optimize everything equally." If all campaigns contribute roughly equally, then improving each by 5% improves the whole by 5%. Time and attention should be spread across the portfolio.
"Diversify across channels." If each channel offers returns drawn from a similar distribution, diversification reduces risk without sacrificing expected value. Being present on six channels is safer than concentrating on two.
"Maximize keyword coverage." If conversions are normally distributed across search queries, comprehensive coverage captures more of the distribution. Missing keywords means missing revenue.
"Judge performance against averages." If outcomes cluster around a mean, then campaigns performing below average are underperforming and need attention. The average itself is a meaningful benchmark.
These practices aren't irrational. Under normality, they're correct. The problem is the underlying assumption.
The Evidence for Power Laws
Power law distributions look different from normal distributions. Instead of outcomes clustering around a mean, they feature a small number of extreme performers and a long tail of minimal contributors. The classic formulation is the Pareto principle (20% of inputs produce 80% of outputs), but power laws are often more extreme than 80/20 suggests.
We see this constantly in performance marketing.
Last quarter we audited a B2B SaaS account. $380K monthly spend. Over a thousand active keywords. The team was proud of their "comprehensive coverage." We pulled the data. Eleven keywords drove 78% of qualified pipeline. Eleven. The rest generated noise, spend, and reports, but almost no business outcomes.
This isn't an outlier. It's the pattern.
Keywords. In most Google Ads accounts, a small fraction of search terms drive the vast majority of conversions. We routinely see 10% of keywords generating 80%+ of conversion volume. The long tail (those hundreds or thousands of terms that seemed worth bidding on) often produces negligible returns relative to spend.
Campaigns. Across accounts, two or three campaigns typically carry the business. The remaining campaigns show activity, but their contribution to actual business outcomes is disproportionately small relative to the budget and attention they consume.
Channels. For most B2B companies, one or two channels drive 90% of results. The third, fourth, and fifth channels (added for "diversification") often dilute focus without proportional return.
Audiences. Within a channel, audience performance is rarely uniform. One segment often converts at 3-5x the rate of others. Spreading budget evenly across segments treats unequal opportunities as equivalent.
Why do averages mislead? Two distributions can have identical means but radically different implications. A normal distribution with a mean of 10 suggests most outcomes cluster around 10. A power law distribution with a mean of 10 might feature a handful of outcomes at 50-100 and a vast majority at 0-2. Managing to the average in the latter case means ignoring the actual structure of returns.
In power law systems, the average is not where the value is. The value is in the tail.
Implications for Budget Allocation
If returns follow a power law, the standard approach to budget allocation destroys value.
In normal distributions, diversification is rational. It reduces variance without sacrificing expected return. In power law distributions, diversification dilutes winners. Capital allocated to the long tail is capital not allocated to the performers that drive actual results.
Consider the typical account review. Budget is spread across campaigns based on historical allocation, category importance, or executive intuition. Campaigns that perform receive similar budgets to campaigns that don't. The implicit logic: each deserves a fair share.
But fair shares make no sense when returns are concentrated.
The rational move is to identify the campaigns driving disproportionate results and concentrate budget there. Potentially dramatically. A campaign generating 50% of conversions at 20% of spend doesn't need proportional budget. It needs more budget until marginal returns actually diminish.
Power law thinking inverts the optimization priority. Instead of asking "how do we improve underperforming campaigns?" the question becomes "are we fully capitalizing on our winners?" The opportunity cost of under-investing in top performers often exceeds the opportunity cost of abandoning the long tail entirely.
One framework that aligns with power law dynamics: the barbell. Allocate most budget to proven winners (high conviction, scale aggressively) and a small portion to cheap experiments (low cost, high potential variance). Minimize the middle, campaigns receiving meaningful budget without demonstrated power law potential.
This feels uncomfortable. It violates the instinct to spread risk, to give everything a fair chance. But if returns are genuinely concentrated, concentration is the correct response.
Implications for Channel Strategy
The same logic applies to channel selection.
The argument for channel diversification assumes each channel offers independent, roughly equivalent return potential. If that's false, if one or two channels drive the vast majority of results, then adding channels dilutes focus, fragments budget, and introduces operational complexity without proportional return.
The question isn't "should we be on TikTok?"
The question is "have we exhausted the power law potential of the channels where we're already winning?"
Usually, the answer is no. There's more headroom in scaling what works than in adding what might.
Every channel requires investment to do well: creative resources, operational attention, platform expertise, testing budget. Spreading these resources across six channels means doing each at one-sixth intensity. If one channel has 10x the return potential of the others, this is a poor trade.
The diversification instinct often comes from risk aversion. "What if Google Ads stops working?" But the risk calculus is wrong. The bigger risk is never achieving full potential in the channel that actually works because resources were diluted across channels that don't.
When should you add a channel? When existing channels are genuinely maxed. When additional budget in your top performer produces diminishing returns. When you've tested every viable audience, structure, and approach and incremental gains are truly incremental.
In practice, this bar is rarely met. The urge to add channels usually precedes the exhaustion of existing opportunity.
Implications for Account Structure
How you structure accounts also encodes distributional assumptions.
The traditional argument for account granularity is that more campaigns and ad groups provide more control. You can set specific bids, allocate specific budgets, and optimize with precision.
But granularity fragments data.
Automated bidding systems (which now drive most sophisticated accounts) need volume to learn. They need enough conversions to identify patterns, enough signal to separate real effects from noise. Fragmenting an account into dozens of campaigns means each campaign has less data, slower learning, and weaker optimization.
In power law terms, granularity prevents the system from identifying and exploiting the power law winners. It enforces a kind of artificial normality.
A caveat here. Granularity makes sense when segments have genuinely different economics. If your California customers have 3x the LTV of Texas customers, you might need separate campaigns with different targets. If one product line has 50% margins and another has 10%, consolidating them confuses the algorithm. Granularity for economic differences: yes. Granularity for the illusion of control: no.
Consolidating campaigns (combining related ad groups, merging audiences, reducing structural fragmentation) enables faster learning and more aggressive optimization of actual winners. The algorithm can see the full distribution and allocate accordingly.
One structural decision (consolidating a fragmented account into a coherent structure) can unlock 2x improvement. One hundred bid adjustments, applied to a fragmented structure, usually can't.
The long tail trap is real. A common audit finding: 85% of spend subsidizes 15% of results. Thousands of keywords, dozens of ad groups, multiple campaigns. All receiving budget, all generating negligible return. The long tail persists because it seems harmless. Each individual keyword costs little. But the aggregate is significant, and the opportunity cost (budget that could scale winners) is invisible.
Cutting the long tail feels risky. What if one of those keywords suddenly performs?
It won't. If it were going to perform, it would have shown signal already. The long tail is where budget goes to die.
Implications for Testing Methodology
Normal distribution thinking shapes testing philosophy in ways that limit value.
Standard optimization focuses on incremental improvements. Adjust bids by 5%. Test a new headline. Tweak audience targeting. Each test might yield a small positive result, accumulating into meaningful gains over time.
But incremental improvements optimize the middle of the distribution. They make average performance slightly better. In a power law system, the middle doesn't matter. A 5% improvement on a campaign generating 3% of results is not a meaningful win.
Smarter testing doesn't mean more testing. It means bigger swings.
Don't test bid adjustments. Test entirely different bid strategies. Don't test headline variations. Test different value propositions. Don't test small audience exclusions. Test entirely different audience approaches.
The goal is to find step-change improvements, not incremental ones. A test that produces no result is fine. It ruled out an option. A test that produces 5% improvement is arguably a waste of time. That effort could have tested something with potential for 50% improvement.
What to test: New bid strategies. New campaign structures. New channel approaches. New audience definitions. Things that could produce 2x improvement if they work.
What not to test: Minor bid adjustments. Small audience exclusions. Incremental tweaks to existing structures. Things that can only produce single-digit improvement at best.
Traditional testing orthodoxy emphasizes statistical significance, enough sample size to be confident in the result. But this framework assumes you're trying to detect small effects. If you're looking for 2x improvements, you need far less data. The signal is obvious.
This doesn't mean abandon rigor. Big swings still need measurement. But waiting months for significance on a test that can only yield 5% improvement is a poor use of time and budget. Save the rigorous measurement for the tests that matter.
Implications for Organization Design
Power law dynamics have organizational implications, mostly around culture and incentives.
Most marketing organizations reward consistency and penalize failure. This makes sense under normality, where failure indicates underperformance. Under power law assumptions, some failure rate is expected. Even optimal. If your team never has tests that fail, they're probably not swinging big enough.
The goal isn't to hire "high variance" people. (How would you even screen for that?) It's to create an environment where good people can take big swings without career risk. Penalizing variance produces safe, mediocre behavior. The marketing equivalent of no one getting fired for buying IBM.
What gets celebrated in your org? A 10% improvement to an underperforming campaign, or finding an entirely new approach that 3x'd results? If the former gets more recognition, you're incentivizing normal distribution behavior.
Reward the identification and scaling of winners. Make "I found something that works, we should pour budget into it" a celebrated outcome, not just "I optimized the portfolio."
Power law dynamics also require the ability to scale winners quickly. When you find something working, can you 5x budget immediately? Do you have the operational capacity and approval processes to capitalize on success?
Most organizations can't. They find a winner and then spend weeks getting approval to increase budget. By the time they scale, the window has shifted. The inability to capitalize on winners is often more costly than the initial failure to find them.
Implementation
Moving from normal distribution thinking to power law thinking requires operational changes.
Start by auditing your current distribution. Look at your actual data through a power law lens. What percentage of spend drives what percentage of results? What percentage of keywords generate what percentage of conversions? The numbers often surprise people.
If 15% of spend drives 60% of conversions, you have a power law distribution. And you're probably under-investing in your winners while subsidizing your losers.
Before looking for new opportunities, find the power law performers in your current account. Which campaigns, keywords, audiences, and channels are already driving disproportionate results? These are your concentration candidates.
Calculate opportunity cost. What would happen if you doubled budget on your top performer? What's the realistic headroom before diminishing returns? Compare that to the expected value of your long tail spend. The opportunity cost of under-concentration is usually larger than people expect.
Build capacity for velocity. Power law marketing requires the ability to move fast. Fast to cut losers. Fast to scale winners. Fast to test big swings. If your approval processes, budget cycles, and operational capacity don't enable speed, you'll identify power law opportunities and fail to capitalize on them.
Shift success metrics. Stop celebrating average improvements. Start tracking distribution shape. What percentage of results come from your top performers? Is that percentage increasing? Are you getting better at identifying and scaling winners?
Summary
Marketing outcomes follow power laws: a small number of inputs generate the majority of results. This reality contradicts the normal distribution assumptions embedded in most marketing operations.
The strategic implications:
The instinct to spread risk, optimize everything equally, and maintain comprehensive coverage is rational under normality. It's irrational under power law dynamics.
Your biggest risk isn't a campaign failing. It's failing to scale your winners.